JPH04314290A - Motion vector detector - Google Patents

Abstract

PURPOSE:To detect a motion vector with accuracy at a pitch being a half one picture element pitch through less calculation quantity and simple constitution. CONSTITUTION:An absolute value (residual difference) of a difference of picture elements at relevant positions between two timewise consecutive frames is stored in a memory 5. A minimum value detection circuit 6 detects a solution vector of one picture element pitch based on the residual difference. An interpolation circuit 9 receives the residual difference corresponding to the solution vector and the residual difference around the vector from the memory 5 to estimate an extreme value with respect to each of longitudinal, lateral and oblique directions. A minimum value detection circuit 10 detects the minimum value in the extreme values and the detected minimum value and the correction value at a half picture element pitch are generated. An adder circuit 7 adds the correction value to the solution vector to obtain a motion vector at a half picture element pitch.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a motion vector detection device used for high-efficiency encoding of moving images, and in particular to a motion vector detection device that generates motion vectors with half-pixel pitch accuracy. Concerning what you do. 2. Description of the Related Art As one method for highly efficient encoding of moving images, differences in pixel data between frames are converted to DCT (Di
A method has been proposed in which the DCT transform output is variable-length coded. In order to efficiently compress moving images, a motion vector detector is provided and the detected motion vectors are also transmitted. Also, to form the inter-frame difference, a local decoder is provided that generates a predicted value of the previous frame. The detected motion vector is provided to this local decoder. For example, block matching is used as a method for detecting motion vectors. FIG. 4A shows a (3×3) block BL0, which is one of the many blocks obtained by dividing the screen of the previous frame. This block B
L0 includes pixel data of B11, B12, . . . , B33. FIG. 4B shows a block AL0 that includes pixel data A11, A12, . . . , A33 within the screen of the current frame and for which a motion vector is to be determined. Within the motion vector search range, block AL of this current frame
The position of 0 is predicted from the previous frame. [0004] First, the absolute value (ABS) (referred to as residual) of the difference between the values of pixels at corresponding positions between two frames in time is calculated, and this residual is added for each block to obtain the residual. The total sum ΔF is calculated. That is, ΔF=A
BS{(A11-B11)+(A12-B12)+...
・+(A33-
B33) A plurality of residual sums ΔF are obtained in the motion vector search range. The minimum value among the plurality of ΔFs is detected, and the position (x, y) of the minimum value is detected as a motion vector. This motion vector (x, y) has a resolution of the distance between pixels (one pixel pitch). [0005] In the above-mentioned high-efficiency encoding method, the higher the accuracy of the motion vector, the less distortion in encoding can be achieved. Therefore, it has been proposed to detect motion vectors with an accuracy of half the pitch of one pixel (referred to as half-pixel pitch). In this method, a predicted block is formed by calculating the average value of two or four points, and block matching processing is performed on this predicted block. [0006] That is, if the corresponding block of the previous frame obtained at one pixel pitch is BL0, then this BL
By calculating the average value of a block that is shifted by one pixel pitch with respect to 0, the following predicted block that is shifted by a half pixel pitch is computed. Prediction block BL1 of (x+0.5,y); {(B11
~B33)+(B12~B34)}/2(x-0.5,
y) prediction block BL2; {(B11 to B33)+(
B10 to B32)}/2 (x+0.5, y+0.5) prediction block BL3; {(B11 to B33)+(B12
~B34)+(B21~B43)+(B22~B44)
}/4 (x-0.5, y+0.5) prediction block BL
4; {(B11-B33)+(B10-B32)+(B
20~B42)+(B21~B43)}/4(x+0.
5, y-0.5) prediction block BL5; {(B11~
B33)+(B12-B34)+(B01-B23)+
(B02-B24)}/4 (x-0.5, y-0.5)
Predicted block BL6; {(B11 to B33)+(B1
0~B32)+(B01~B23)+(B00~B22
)}/4(x,y+0.5) prediction block BL7; {
(B11-B33)+(B21-B43)}/2(x,
y-0.5) prediction block BL8; {(B11 to B3
3)+(B01-B23)}/2 [0007] Block AL0 of the current frame and predicted block BLi (i=0,1
, 2, . . . , 8) is detected, and the vector for this detected prediction block Bi is taken as a motion vector with half pixel pitch accuracy. The calculation amount of this method for determining a half-pixel pitch motion vector is as follows. Here, the size of one block is (N×N
), and the amount of calculations after the motion vector of one pixel pitch is determined is investigated. First, if the prediction block calculation is limited to addition only, (addition for prediction of 2-point average) + (addition for prediction of 4-point average) = 4N2 + 6 × 2N2 = 16N2 [
[0009] In order to obtain the absolute value of the difference, as described above, it is necessary to perform 4N2 subtraction and 4N2 absolute value calculation for each of the two-point average prediction and the four-point average prediction. Subtraction) + (absolute value) = (8N2) + (8N2)
=16N2 [Problem to be Solved by the Invention] In order to obtain the sum of absolute value differences for each block, (
Since the addition of (N2 - 1) is required, the total number of additions is (8N
2-8) calculations are required. Actually, (N=16
), and in that case, 4096 additions, 2048 subtractions and absolute values, and 2040 additions are required per block. Furthermore, including the time required to access each pixel data from memory, it is difficult to process it in real time. When adopting a hardware configuration such as parallel processing for high-speed processing, a problem arises in that the scale of the hardware increases. [0011] Accordingly, an object of the present invention is to provide a motion vector detection device capable of detecting a half pixel pitch motion vector with fewer calculations than the conventional method. Means for Solving the Problems The present invention obtains a residual, which is the absolute value of the difference in pixel values at corresponding positions between two temporally consecutive frames, and calculates this residual for each block. and means (3, 4, 6) to calculate the solution vector of the block with an accuracy of 1 pixel pitch from the minimum value of this total value, and perform interpolation using the residual error around the solution vector of 1 pixel pitch. By doing this, in each of multiple directions, the extreme value of the half pixel pitch is estimated, the minimum value among the extreme values is detected,
A motion vector detection device comprising means (9, 10) for generating a correction value corresponding to the minimum value, and means (7) for generating an output vector in which a correction value is added to a solution vector of one pixel pitch. be. [Operation] A solution vector of one pixel pitch is detected. The residuals around this solution vector are used to find extrema in multiple directions by interpolation. The minimum value among the plurality of extreme values is detected and used as a correction value. By adding a correction value to the one-pixel pitch solution vector, a half-pixel pitch motion vector is obtained. [Embodiment] An embodiment of the present invention will be described below with reference to the drawings. Motion vector detection in an embodiment of the present invention involves interpolating and estimating the extreme values of the half pixel pitch in multiple directions from the residuals around the solution vector determined by the pixel pitch, and then setting the minimum value of these extreme values as a correction value. By adding to the solution vector of one pixel pitch, a motion vector with half pixel pitch accuracy is obtained. FIG. 1 shows a schematic block diagram of one embodiment of the invention. Digital video data is input to an input terminal indicated by 1 in FIG. Each frame of this input video data is alternately stored in frame memories 2A and 2B by switching circuit SW. Pixel data read out from the previous frame stored in the frame memory 2A and the current frame stored in the frame memory 2B are supplied to the subtraction circuit 3. The frame difference for each pixel from the subtraction circuit 3 is supplied to an absolute value converting circuit 4, where it is converted into an absolute value. The absolute value of this frame difference, that is, the residual difference, is stored in the memory 5. The residual error stored in the memory 5 is detected by the minimum value detection circuit 6.
is supplied to The minimum value detection circuit 6 detects a solution vector of one pixel pitch by block matching. This method is similar to motion vector detection using conventional block matching. If the motion search range is M both vertically and horizontally, from the position (-M, -M) to (+M, +
The minimum value among the residuals for the range of M) is detected as the solution vector. This solution vector is supplied to memory 5 and adder circuit 7. A plurality of residuals around the above solution vector are read out from the memory 5. Figure 2 shows the solution vector (x
, y) of (3×3=9) residuals D00 to D
22 is shown. The residuals D00 to D22 are supplied to the interpolation circuit 9, and the interpolation circuit 9 estimates extreme values in four directions. The interpolation operation of the interpolation circuit 9 will be explained with reference to FIG. In FIG. 3A, L1, L2, L3, L
4 represent horizontal, vertical, and diagonal interpolation directions, respectively. For example, regarding the diagonal direction L4, the residual D0
2, D11, and D20 exist. As shown in FIG. 3B, where the horizontal axis is the position in the direction of L4 and the vertical axis is the value V, if D20>D02, then d02=D02-{(
D20-D11)/2}, the extreme value d02 of the half pixel pitch in the direction of L4 is estimated. If D20≦D02, d21=D20−{(D
02-D11)/2}, the extreme value d21 of the half pixel pitch in the direction of L4 is estimated. The V-shaped interpolation calculation as shown in FIG. 3B is performed in the other direction L1.
, L2, and L3, and the estimated extreme values in each direction are detected. The extreme values of the four half-pixel pitches estimated by the interpolation circuit 9 are supplied to the minimum value detection circuit 10. The minimum value detection circuit 10 detects the minimum value among these four extreme values by comparing the magnitudes, and the detected extreme value is D1.
When the condition smaller than 1 is satisfied, the value corresponding to the minimum value is adopted as the half pixel pitch correction value. If the detected extreme value is greater than or equal to D11, no correction value is generated (that is, the correction value is 0). As an example, by the interpolation circuit 9, d00, d01, d02, d in FIG.
10 is estimated as the extreme value, d00 is detected as the minimum value by the minimum value detection circuit 10, and furthermore, d0
If 0<D11, a correction value of (-0.5, -0.5) is generated. The correction value from the minimum value detection circuit 10 is supplied to the addition circuit 7. In the addition circuit 7, a correction value is added to the pixel pitch solution vector (x, y),
At the output terminal 8, a motion vector with half-pixel pitch accuracy can be obtained. In this embodiment, a half pixel pitch motion vector can be calculated for one block by performing four value comparisons, eight subtractions, four shift divisions, and minimum value determination. In order to obtain a solution vector of one pixel pitch, various techniques may be used to reduce the amount of calculation, such as a three-step search method. Further, the interpolation method for finding extrema in a plurality of directions using residuals around the solution vector is not limited to the one described in the above embodiment. [0022] According to the present invention, there is no need to form a plurality of prediction blocks at a half-pixel pitch as in the conventional method, so that the amount of calculation can be significantly reduced compared to the conventional method. Further, the present invention has the advantage of a simple circuit configuration because a simple interpolation calculation is performed using the previously determined residual and a correction value is detected from the minimum value.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an embodiment of the present invention.

FIG. 2 is a schematic diagram used to explain the operation of an embodiment of the present invention.

FIG. 3 is a schematic diagram used to explain the operation of an embodiment of the present invention.

FIG. 4 is a schematic diagram for explaining conventional motion vector detection.

[Explanation of symbols]

2A, 2B Frame memory 5 Memory for storing residuals

Claims (1)

[Claims] Claim 1: Obtain a residual, which is the absolute value of the difference in pixel values at corresponding positions between two temporally consecutive frames, sum this residual for each block, and calculate from the minimum value of this total. By performing interpolation using the means for determining the solution vector of the block with an accuracy of one pixel pitch and the residual error around the solution vector with one pixel pitch, half pixel pitch can be obtained in each of a plurality of directions. means for estimating an extreme value, detecting a minimum value among the extreme values, and generating a correction value corresponding to the minimum value; and an output in which the correction value is added to the solution vector of one pixel pitch. A motion vector detection device comprising means for generating a vector.